Merge branch 'main' into liyang/upcast_mxfp_and_dot_scaled

Author: etiotto

.github/workflows/auto-update-translator-cid.yml

@@ -66,7 +66,7 @@ jobs:
 
       - name: Get commit ID from Triton's spirv-llvm-translator.conf
         if: ${{ env.TARGET_PRID == null }}
-        run: echo "CID_CURRENT=$(cat ./lib/Target/SPIRV/spirv-llvm-translator.conf)" >> $GITHUB_ENV
+        run: echo "CID_CURRENT=$(<third_party/intel/lib/Target/SPIRV/spirv-llvm-translator.conf)" >> $GITHUB_ENV
 
       - name: Checkout SPIRV-LLVM-Translator
         if: ${{ env.TARGET_PRID == null }}
@@ -82,12 +82,12 @@ jobs:
         run: |
           echo "CID_LATEST=$(git -C external/SPIRV-LLVM-Translator rev-parse HEAD)" >> $GITHUB_ENV
 
-      # the latest valid cid has been weitten to spirv-llvm-translator.conf
+      # the latest valid cid has been written to spirv-llvm-translator.conf
       - name: Search the latest valid Translator cid
         if: ${{ env.TARGET_PRID == null }}
         run: |
           ./scripts/check-update-translator-cid.sh $CID_LATEST $CID_CURRENT
-          if git status --porcelain ./lib/Target/SPIRV/spirv-llvm-translator.conf | grep '^ M'; then
+          if git status --porcelain third_party/intel/lib/Target/SPIRV/spirv-llvm-translator.conf | grep '^ M'; then
             echo "MODIFIED=true" >> $GITHUB_ENV
             echo "spirv-llvm-translator.conf has been modified"
           fi
@@ -112,7 +112,7 @@ jobs:
           git checkout -b ${PR_BRANCH}
           git branch --show-current # bot/update_translator_cid
           git status
-          git add ./lib/Target/SPIRV/spirv-llvm-translator.conf
+          git add third_party/intel/lib/Target/SPIRV/spirv-llvm-translator.conf
           git commit -m "Update spirv-llvm-translator.conf"
           git push origin ${PR_BRANCH}
-          pr_url=$(gh pr create --title "[github-bot] Update spirv-llvm-translator.conf" --body "Automated PR to update translator commit id." --reviewer whitneywhtsang --head ${PR_BRANCH} --base main)
+          gh pr create --title "[github-bot] Update spirv-llvm-translator.conf" --body "Automated PR to update translator commit id." --reviewer whitneywhtsang --head ${PR_BRANCH} --base main

.github/workflows/triton-benchmarks.yml

@@ -202,6 +202,7 @@ jobs:
           python gemm_streamk_benchmark.py --reports $REPORTS
           source ../../scripts/capture-hw-details.sh
           python ../../scripts/build_report.py $REPORTS/matmul-streamk-performance.csv $REPORTS/gemm-streamk-triton-report.csv --benchmark gemm-streamk --compiler triton --param_cols "M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
+          python ../../scripts/build_report.py $REPORTS/matmul-streamk-performance.csv $REPORTS/gemm-streamk-xetla-report.csv --benchmark gemm-streamk --compiler xetla --param_cols "M,K,N" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
 
       - name: Run Triton GEMM (split-k) kernel benchmark
         if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_splitk_benchmark.py') }}
@@ -210,6 +211,7 @@ jobs:
           python gemm_splitk_benchmark.py --reports $REPORTS
           source ../../scripts/capture-hw-details.sh
           python ../../scripts/build_report.py $REPORTS/matmul-splitk-performance.csv $REPORTS/gemm-splitk-triton-report.csv --benchmark gemm-splitk --compiler triton --param_cols "M,K,N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
+          python ../../scripts/build_report.py $REPORTS/matmul-splitk-performance.csv $REPORTS/gemm-splitk-xetla-report.csv --benchmark gemm-splitk --compiler xetla --param_cols "M,K,N" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
 
       - name: Run Triton GEMM + PreOp (exp) kernel benchmark
         if: ${{ steps.install.outcome == 'success' && !cancelled() && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_preop_exp_benchmark.py') }}

include/triton/Dialect/Triton/IR/TritonTypes.td

@@ -15,7 +15,7 @@ class TritonTypeDef<string name, string _mnemonic, list<Trait> traits = []>
 }
 
 // Floating-point Type
-def TT_Float : AnyTypeOf<[F8E4M3FN, F8E4M3FNUZ, F8E4M3B11FNUZ, F8E5M2, F8E5M2FNUZ, F16, BF16, F32, F64], "floating-point">;
+def TT_Float : AnyTypeOf<[F8E4M3FN, F8E4M3FNUZ, F8E5M2, F8E5M2FNUZ, F16, BF16, F32, F64], "floating-point">;
 def TT_FloatTensor : RankedTensorOf<[TT_Float]>;
 def TT_FloatLike : AnyTypeOf<[TT_Float, TT_FloatTensor]>;
 

python/src/ir.cc

@@ -807,11 +807,6 @@ void init_triton_ir(py::module &&m) {
            [](TritonOpBuilder &self) -> Type {
              return self.getBuilder().getI8Type();
            })
-      .def("get_fp8e4m3b11fnuz_ty",
-           [](TritonOpBuilder &self) -> Type {
-             // TODO: align with upstream code to use i8
-             return self.getBuilder().getType<Float8E4M3B11FNUZType>();
-           })
       .def("get_fp8e5_ty",
            [](TritonOpBuilder &self) -> Type {
              return self.getBuilder().getType<Float8E5M2Type>();
@@ -1680,11 +1675,11 @@ void init_triton_ir(py::module &&m) {
              if (haveDiagnostics) {
                context->printOpOnDiagnostic(true);
                context->printStackTraceOnDiagnostic(true);
+               context->getDiagEngine().registerHandler([](Diagnostic &diag) {
+                 llvm::outs() << diag << "\n";
+                 return success();
+               });
              }
-             context->getDiagEngine().registerHandler([](Diagnostic &diag) {
-               llvm::outs() << diag << "\n";
-               return success();
-             });
              if (haveDump) {
                auto printingFlags = OpPrintingFlags();
                printingFlags.elideLargeElementsAttrs(16);

scripts/check-update-translator-cid.sh

@@ -25,7 +25,7 @@ COMMIT_IDS=$(git -C $TRITON_PROJ/external/SPIRV-LLVM-Translator log --format="%H
 cd $TRITON_PROJ
 FOUND=false
 for cid in $COMMIT_IDS; do
-    echo "$cid" > ./lib/Target/SPIRV/spirv-llvm-translator.conf
+    echo "$cid" > third_party/intel/lib/Target/SPIRV/spirv-llvm-translator.conf
 
     BUILD_STATUS=PASS
     echo "::group::Building Triton for $cid"
@@ -52,5 +52,5 @@ for cid in $COMMIT_IDS; do
 done
 
 if [ "$FOUND" = false ]; then
-    git restore ./lib/Target/SPIRV/spirv-llvm-translator.conf
+    git restore third_party/intel/lib/Target/SPIRV/spirv-llvm-translator.conf
 fi

third_party/intel/language/intel/utils.py

@@ -25,7 +25,7 @@ def num_warps(_builder=None):
 
 def convert_fp8e4b15_to_float16(arg, _builder):
     # Need to bitcast the source first because it's represented as tensor of i8 in MLIR.
-    tmp_ty = _builder.get_block_ty(_builder.get_fp8e4m3b11fnuz_ty(), arg.type.shape)
+    tmp_ty = _builder.get_block_ty(_builder.get_fp8e4b8_ty(), arg.type.shape)
     tmp = _builder.create_bitcast(arg.handle, tmp_ty)
     # Now generate FpToFp op for upcast.
     dst_ty = core.block_type(core.float16, arg.type.get_block_shapes())

third_party/intel/lib/Target/SPIRV/spirv-llvm-translator.conf

@@ -1 +1 @@
-c1828a93d050afe8a44cc239a2908270c578c345
+50af2c70832745be1c0ce1562c4038162c319e5c

third_party/intel/lib/TritonIntelGPUToLLVM/ElementwiseOpToLLVM.cpp

@@ -974,7 +974,7 @@ struct FpToFpOpConversion
   std::pair<ConverterT, size_t>
   getConversionFunc(Type srcTy, Type dstTy,
                     std::optional<RoundingMode> roundingMode) const {
-    auto F8E4M3B15TyID = TypeID::get<Float8E4M3B11FNUZType>();
+    auto F8E4M3B15TyID = TypeID::get<Float8E4M3FNUZType>();
     auto F8E4M3TyID = TypeID::get<Float8E4M3FNType>();
     auto F8E5M2TyID = TypeID::get<Float8E5M2Type>();
     auto F16TyID = TypeID::get<Float16Type>();